Also this week: why the AI trade only cracks on demand, an investment committee fight over Zipline, fal and Revolut, the API key crime wave, and why seat-based B2B can’t get a bid
This was one of the busiest news weeks we’ve had on the pod. Sequoia unwound years of expansion in a single memo. Nvidia’s board opened an investigation into its founder, and he resigned the day after we recorded. The NASDAQ 100 lost $2.5 trillion in a day. Anthropic set a conservative IPO range, and the book is now reportedly about six times oversubscribed. And agents kept pulling growth away from seat-based software. Harry Stebbings, Rory O’Driscoll and I went through all of it.
1. Sequoia Is Winding Down the Evergreen Fund. That’s the Real Story.
Sequoia told its LPs on November 4 that it is closing its London office, shutting its three-year-old Arc accelerator and ending its scout program, while raising about $10 billion in new funds. It is also merging early and growth into one team and winding down its evergreen structure, which had pushed LPs to keep holding positions long after the IPO.
Jason’s take: The partner moves and London are the least interesting part. What I want to know is what happened to the evergreen idea. Sequoia owned the dominant brand in venture and had a structure built to hold generational assets for decades. Picture still owning all their Nvidia and all their Google. That was the best idea in venture, and they’re walking away from it.
Rory’s read: The thinking was sound, since the biggest returns come after the IPO. It ran into three problems. It launched in 2022-23, right before stocks fell 30%. LPs have different liquidity needs. And partners who have cleared carry can already hold their own stock forever, so forcing LPs to hold amounts to “giving the LPs their vitamins.” Rory’s example: a Sequoia partner has sat on Nvidia’s board since 1993 and is probably worth billions from holding. That was his call to make. It doesn’t mean every LP should have been locked into the same trade.
The founder lesson: The firm may last. No individual partner will. Founders tend to treat their VCs as more permanent than they are.
2. Sequoia Thinks It Can Cover Europe From California. It’s Probably Right Above Pre-Seed.
Sequoia is closing its London office six years after opening it, and several people will leave the firm. Andreessen Horowitz closed its London office in 2025 and Coatue pulled out in 2024.
Harry’s read: Sequoia did well in Europe with ElevenLabs, Dollar App and Bolt. The message is that it can cover those founders from the US.
Jason’s take: Mati from ElevenLabs is at every top-tier dinner I go to in the Bay. If it weren’t for the accent, I’d assume ElevenLabs was a Bay Area company. The best European founders are here every month, and Sequoia’s brand gets it a chance to put a term sheet in front of them. Pre-seed is the exception. That still needs someone local doing the hands-on work.
On killing Scouts and Arc: Rory pointed out scouts worked best while the program was secret, and it stopped being secret once everyone copied it. Jason read it as focus. A tight partnership doesn’t want to run a scout network plus an accelerator that sits below Speedrun, let alone YC. He once looked at running a SaaStr accelerator, saw it meant hiring and managing 15 people, and passed.
The Founders Fund comparison: Harry noted Founders Fund deployed its growth fund in 11 months with a handful of partners, into Stripe, Anduril and SpaceX. Rory’s rule: spread out early, kill the losers fast, then concentrate hard. Founders Fund went from about 20 positions in the early fund to seven or eight in growth.
3. Jensen Resigned the Day After We Recorded. Rory Called It.
Update: We recorded on Tuesday, November 10. On November 11, Nvidia’s board accepted Jensen Huang’s resignation as President, CEO and director, effective immediately, and appointed CFO Colette Kress interim CEO while it searches for a permanent one. Harvey C. Jones was named the board’s first-ever chair, and Nvidia will still report Q3 on Tuesday, November 17, as scheduled. The independent directors plan to keep investigating despite the resignation.
The background: on November 8 the board said a committee of independent directors with outside counsel would investigate reporting that Huang was the target of a foreign intelligence effort. Huang had acknowledged to the board an intimate relationship with a business executive whom U.S. intelligence reportedly determined was gathering information for Beijing. She was an executive at Megaspeed, a Singapore-based Nvidia customer.
What we said on Tuesday. Jason argued Jensen needed a pass. He had run Nvidia for 33 years, built a $5 trillion company, and mattered too much to AI and to the US, the way Apple needed Steve Jobs during the options backdating scandal. That didn’t hold up. Rory’s answer did: “You can hope a pass, but you can’t assume it.” The probe would cover compliance, Megaspeed and how Nvidia dealt with Washington, and there was a nonzero chance something bad would surface. Two days later, the board decided it couldn’t wait for the findings.
Rory’s irony, which still holds: Jensen didn’t need a spy to persuade him to sell chips to China. He has argued for selling to anyone, consistently and in public, because that’s what is good for Nvidia.
Jason’s call, which also still holds: China loses. Selling GPUs to China gets harder, and so does parking them in data centers just over the border.
The governance lesson: Jason has been through two board investigations of CEOs. His point was that who you hire to investigate shapes the result. Rory’s counter was that scrutiny changes behavior. When Congress and the DOJ will check your homework, even a long-tenured, loyal board can’t look away. This board acted before its own investigation finished. And for anyone keeping score from section 1: two days after we said the founder CEO usually lasts longest in tech, the longest-serving one was out.
4. NASDAQ 100 Loses $2.5 Trillion in a Day. The AI Trade Only Cracks on Demand.
Harry opened with hedge fund friends calling him more down than he’d heard them in years.
Rory’s read: Nobody knows anything. The market rose after Tuesday’s election, rose again Wednesday, then fell Thursday on economic data and worries about data center buildouts. The AI trade doesn’t crack on supply. Not because data centers get blocked, and not because Nvidia loses share to AMD. It cracks only if demand cracks, and the demand will show up in OpenAI’s and Anthropic’s ARR. Last Thursday was overextended hedge fund pods getting stopped out. Shops like Citadel cut your risk at a 5% drawdown and fire you at 7.5%.
Jason’s add: Some nervousness is fair. This buildout needs an enormous amount of capital and debt.
5. Anthropic Set a Conservative Range. The Book Is Now About 6x Oversubscribed.
Update: The latest prospectus sets a price range of $70 to $78 per share, with pricing expected November 18 and trading November 19. Seven compute suppliers, including Nvidia, Google, AMD, Broadcom, Micron, Microsoft and TSMC, have agreed to buy $26 billion of non-voting stock at the IPO price. Since we recorded, Bloomberg reported the IPO has drawn about six times the orders for the shares available, with existing backers getting priority and some buyers receiving smaller allocations than they expect. The range isn’t the price. With a book that deep, Anthropic could raise the range before it prices. Or it could hold the range and price at the top. Nobody knows yet.
Rory’s read, which held up: When I said on Tuesday that the round didn’t look filled, Rory pushed back. The book wasn’t at the oversubscription stage yet. He saw the range as a deliberate starting point: generate interest, then walk it up. His line was that in an IPO you would much rather start low and move up than start high and come back down. He also expected Anthropic would either raise the filing range or at least price at the high end and declare victory. He also explained the $1 trillion vs. $1.4 trillion confusion. Private valuations count all granted options. Public market caps count only vested ones.
Jason’s read, which didn’t: I said the market was fatigued and called it flat, partly because the private placement looked hard to fill. A 6x book says demand going in is anything but fatigued. We’ll find out on November 18 whether that turns into a higher range or a higher price, and on November 19 how it trades.
Harry’s data point still matters: One of the largest trillion-dollar allocators told him they aren’t worried about this $80 billion. They’re worried about the next $100 billion, because that layer of capital is very thin. Six times demand on this deal doesn’t answer that question.
Rory, the bull: This is the first chance to buy a pure-play generative AI company rather than picks and shovels. And if growth slows, Anthropic will have roughly $200 billion in cash, more than any other company in America.
6. You’d Rather Be Anthropic Than OpenAI. OpenAI Is Still Reaccelerating.
Greg Brockman said OpenAI’s annualized revenue grew 70% in Q3 2026 and that enterprise now generates the majority of revenue, after a first half he called “pretty rough”.
Jason’s take: Anthropic will have the cash and the float, and it won’t need to go back to private investors. Expect a convert within six to nine months and another $50 billion raised using every tool a big public company has. OpenAI is accelerating, but it is still private, unprofitable and raising money in pieces. And quarterly numbers from Anthropic are a gift to everyone who has been piecing this market together from CoreWeave, Nvidia and Nebius filings.
Rory’s read: If OpenAI really reaccelerated to 70%, that’s stunning, and faster than Anthropic for the quarter. The real question now is how price-sensitive token demand is. If a 20% price cut brings 21% more usage, revenue goes up. If it brings 15% more, revenue goes down. Competition comes down to compute, capital and token pricing.
On consumer agents: OpenAI also said its Dots agent product is expensive and compute-heavy. Jason’s view: Lovable and Replit cost a few dollars per user because they run on short-lived VMs. A persistent agent costs closer to $20 to $30 a month, and if it can’t run 24/7 it’s just an automated workflow. Rory: compute costs fall eventually, but that could be two years away or five, and he guesses five. Harry: Meta can afford to wait.
7. Investment Committee: Zipline vs. fal vs. Revolut
- Zipline: raised $1.5 billion at a $21.5 billion post-money valuation, with Walmart signing a national deal for 300 additional sites. That’s 2.8 times its January valuation of $7.6 billion.
- fal: raised $1.3 billion at a $16 billion valuation, up from $8 billion in April, about 20 times the $800 million in annualized revenue a source cited to Bloomberg.
- Revolut: a secondary sale valued it at $115 billion, up 53 percent from $75 billion in November 2025.
Jason’s pick: fal. This was the first week he’d be happy doing all three. Zipline now looks like a winner with better unit economics than Waymo. Revolut is the simplest double or triple. He still picks fal, while knowing compute becomes a commodity in three to five years and the margins are thin. Sometimes in venture you have to be a tourist and ride the wave. His worst case is flipping it at $40 billion.
Harry’s pick: Revolut. He’s getting uneasy about this market and wants safety. Revolut is the best asset on the list with the best founder of our generation. Buy at $115 billion, compound for 10 years, call it a day. His one risk is Nik: “I want him near no roads, cars, any moving objects.”
Rory’s read: On Zipline, it’s roughly 3 million US deliveries at maybe $5 to $10 each. At $21.5 billion you are betting heavily on the Walmart contract. On fal, the third-party inference market is around $5 billion and growing fast. He worried too much about the structure of that market early on, and was wrong, because when demand triples or quadruples every year nobody cares about the long run. But at $16 billion you may be closer to the end of that run than the start. So he joins Harry on Revolut. Its 19-to-25-year-old customers will want mortgages over the next 20 years, which puts Revolut in position to become Europe’s biggest bank.
8. Stolen AI API Keys Are the Crime of the Year
Fireworks, valued at $17.5 billion in July, had two major breaches in 30 days.
Jason’s take: Bot swarms are going after every Anthropic and OpenAI key they can find. Railway got hit. Others got hit. Jason found his own Claude spending limit had defaulted to $100,000, and a stolen key can burn through that in minutes. “This is better than selling Rolexes you steal off someone’s wrist in London.” The awkward part: customers like Harvey can’t leave, because the capacity doesn’t exist anywhere else.
Harry’s data: Wiz said nine leading AI developers disclosed swarming incidents in October alone. Unit 42 says bills reach $1 million before anyone notices.
The M&A angle (Rory): CrowdStrike and Palo Alto keep buying agent security startups for around $400 million each, and Meta and OpenAI are buying coding tools as the software development lifecycle gets rebuilt from scratch. When a new problem becomes the top issue, the platforms buy their way in.
9. Dialpad Buys Ada. That’s Not an Exit.
Dialpad agreed to acquire Ada in an all-share deal valuing it at about $430 million. Ada raised a $130 million Series C in May 2021 at a reported $1.2 billion valuation.
Rory’s read (disclosure: he’s a Dialpad investor): Ada was a good chatbot company from before LLMs. Even after rebuilding on them, it had lost its momentum, and competing with Sierra and Decagon is hard. Joining a bigger company makes sense.
Jason’s take: No liquidity means it isn’t really an exit. What bothers him more is that going into 2027, so many pre-AI companies still ship at their 2021 pace. Canva is too slow. Board updates still promise “an AI-native team next year,” which was the 2025 plan. For portfolio companies, December 31 may be the point where you give up. And the AI-native companies won’t buy the slow ones for their customers, because they’re growing too fast for it to be worth it.
Harry’s suggestion: Thoma Bravo should pay Jason $50 million a year to show its CEOs how to go AI-first. Rory’s version: have him depress the companies Orlando wants to buy so they sell at 3x run rate.
Rory’s close: It’s march or die. That was a hot take two years ago. Now it’s a given.
10. Cloudflare Grows 39% Because of Agents. Atlassian Falls on Seats.
Cloudflare’s Q3 revenue was $782.5 million, up 39%, and it crossed $3 billion in annualized revenue, six months after cutting about 20% of its workforce for an “agentic AI-first operating model”. Harry also flagged Freshworks crossing $1 billion with 30% fewer employees, Palantir at about $2 million in revenue per employee, and Atlassian dropping after warning on seat and subscription growth.
Jason’s take: AI is partly the excuse. Most of these companies never needed that many people, and AI just made it obvious. Agents specifically, not LLMs in general, are what took Cloudflare from the 20s to 39% growth and what’s driving agentic spend at Twilio. For 2027, if a company isn’t benefiting from agents, he doesn’t want to hear about it. Agents don’t care about your UX, and humans are a shrinking share of every user base, even in vertical B2B.
Rory’s read: With a huge demand tailwind, you don’t need the marginal 2021 tactics. You staff the machine and take the business as it comes. With $2 to $5 million contracts like Palantir’s, you can’t hire fast enough to keep up, so margins rise whether you want them to or not. For seat-based software, the trend has turned against you.
11. Figma Below $10 Billion: Turning Down an Exit Has a Long Tail
Figma fell below $10 billion in market cap this week.
Jason’s take: It would eat at him. Figma had a $20 billion-plus Adobe offer with instant liquidity in 2022 until regulators blocked it, and he worries antitrust gets tougher again with the new Congress. The market has decided seat-based models like Figma, Monday and HubSpot are terminal. He doesn’t think that’s literally true, but he expects agent-driven work to keep pulling the money toward Cloudflare and Twilio.
Rory’s read: Figma is still growing around 40% with strong free cash flow, and the recent hit came from spending more on AI. But it proves the boring M&A advice. Turn down X, execute brilliantly for three years, go public, and you can still end up worth half of X. Jason added the part nobody talks about: once the stock is down, a malaise sets in, top performers drift, and recruiting gets harder.
If Jason were Figma’s CEO: Go all in on agentic design. Today, if you ask Claude, Replit or Lovable for a fully designed end-to-end app, you don’t get one. Their design modes are a C-minus. Figma should promise every page of your app at Notion-level quality, built by its own agents, and accept the gross margin hit. Otherwise it ends up like Adobe: still in the workflow, but agents go around it. Rory agreed. His team watched designers do the same task in Canva, Figma and newer agentic tools, and the agentic tools were simply faster.
12. California Rejects the Billionaire Tax, in a Big Blue Year
Voters rejected Proposition 40, a tax on billionaires’ net worth, by 52.8% to 47.2%. It would have imposed a one-time 5% tax on about 200 Californians worth more than $1 billion. Proposition 42, which limits new taxes on personal holdings, passed with 52%. Sergey Brin led the opposition with $102 million to a committee he founded.
Rory’s read: The contrast is what’s interesting. The country swung Democratic harder than at almost any point since the 1970s, and the most Democratic state still said no. California voters already lean no on propositions, and the No side won the argument even though nobody in 2026 is cheering for tech billionaires. Voters sent a clear message to Washington without wrecking things at home.
Jason’s take: The No side outspent the unions roughly 10 to 1, which matters most on ballot measures about money. Still, this outcome was important. He spoke with two quiet billionaire founders who were already planning to leave. These are people building companies, not passive investors. If it had passed, it would have told the most successful founders California is hostile, and that would have spread. By Jason’s math, Anthropic’s IPO alone will create about $41 million per employee on average. The company may stay put, but employees can move when they make their tax decisions.
Top Quotable Moments
Harry Stebbings
- “Give me Revolut at 115 every day. Let me compound for 10 years and I’m calling it a day.”
- On Nik Storonsky: “Do not let him leave the building. I want him near no roads, cars, any moving objects.”
- “If I was Orlando Bravo, I would be paying 50 million bucks a year to Mr. Jason Lemkin to come in and talk to our CEOs about how to make our very monolithic, old, stale organizations AI first.”
Rory O’Driscoll
- “It is far better in an IPO to start low and go up than to start high and come back, because once you start coming back everyone loses confidence and you’ve blown it.”
- “You could work for the next three years, do an amazing job, grow your damn company 40% a year, get the damn thing public. And at the end of that, you’re going to be worth half what you are today. No one ever said capitalism was fair or easy.”
- “You can hope a pass, but you can’t assume it.”
Jason Lemkin
- “You’d much prefer to be these guys than OpenAI right now. They’re going to have the money. They’re going to have the float.”
- “This is better than selling Rolexes you steal off someone’s wrist in London. This is the crime of the year.”
- “Agents don’t care about your UX. Humans do, but they’re a declining percent of everybody’s user base.”
